cuda not recognizing driver in Linux Mint 17.2 (Ubuntu 14.04)

I am trying to install cuda for a geforce 960 in a machine running ubuntu 14.04 ***edit: Linux Mint 17.2

After following all the directions in ‘getting started’ guide I get the following error when running
$ sudo apt-get install cuda

Reading package lists… Done
Building dependency tree
Reading state information… Done
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
cuda : Depends: cuda-7-5 (= 7.5-18) but it is not going to be installed
E: Unable to correct problems, you have held broken packages.

If I try to install cuda-7-5 and then cuda-runtime-7-5 I finally come up on the error:

The following packages have unmet dependencies:
cuda-runtime-7-5 : Depends: cuda-drivers (>= 352.39) but it is not going to be installed
E: Unable to correct problems, you have held broken packages.

This seems to say to me that the package thinks I have an older driver. However, when I check I have driver 352.41. Does anybody know what might be going on here?

I’m not one to do the “me too” threads, but I am having the same problem. If you do find an answer, please let me/us know.

Same problem here on Linux Mint 17.2 (which is basically equivalent to Ubuntu 14.04). I suspect there’s a bug in the repository, because I get that error whether I try to install cuda-7-5, cuda-7-0, or cuda-6-5. It doesn’t seem to matter whether I use the 7.0 or 7.5 .deb file, or whether I use the local or network installer. Anything using apt-get seems to be broken. I also tried installing via the .run file, but I was unsuccessful at getting Nouveau to turn off and stay off (unsurprisingly). If the .run file doesn’t try to pull cuda-drivers from the repo, then fighting to the death with Nouveau might be the only option until nVidia fixes the other installation methods.

[s]I have an update. First, I am actually using Mint 17.2 (which I also thought was basically 14.04, but maybe not…) Using the .run file I was able to get a successful installation and was able to run deviceQuery once. I followed directions here: https://www.reddit.com/r/deepdream/comments/3cd1yf/howto_install_on_ubuntulinux_mint_including_cuda/

(screenshot of successful deviceQuery attached)

Sadly, after trying to run some code using the torch packages “cutorch” and “cunn” I received this error:

package cunn not found!
package cutorch not found!
If cutorch and cunn are installed, your CUDA toolkit may be improperly configured.
Check your CUDA toolkit installation, rebuild cutorch and cunn, and try again.
Falling back on CPU mode

When I went back to run deviceQuery as a sanity check, I no longer got information about my device, but the following error:

./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 35
→ CUDA driver version is insufficient for CUDA runtime version
Result = FAIL

not sure what is going on…[/s][/s][/s]

I have an update!
First, I am actually using Mint 17.2 (which I also thought was basically 14.04, but maybe not…)

Using the .run file I was able to get a successful installation and was able to run deviceQuery once. I followed directions here: https://www.reddit.com/r/deepdream/comments/3cd1yf/howto_install_on_ubuntulinux_mint_including_cuda/

CUDA appears to be working correctly! Im still having some trouble getting it to interface with torch, and for graphics based samples it is being fussy - but it is installed and functioning!

A note: properly uninstall the failed CUDA before re-doing with the above instructions.

“https://www.reddit.com/r/deepdream/comments/3cd1yf/howto_install_on_ubuntulinux_mint_including_cuda/”

Thanks for finding that! It worked! I had to make one small change for the CUDA setup: After successfully installing the nvidia driver, cuda library, and cuda samples, I had to reinstall NVIDIA driver 352:

sudo add-apt-repository ppa:xorg-edgers/ppa
sudo apt-get update
sudo apt-get install nvidia-352 nvidia-prime
sudo add-apt-repository -r ppa:xorg-edgers/ppa

Then I was able to log in, and run:

cd /usr/local/cuda/samples/1_Utilities/deviceQuery
sudo make
sudo ./deviceQuery

For a success! Thanks again!

There is a relatively simple solution. Download the package nvidia-settings_352.39-0ubuntu1_amd64.deb from the nvidia repository and install it using
dpkg -i nvidia-settings_352.39-0ubuntu1_amd64.deb
After that, the rest of cuda tools can be installed using the recommended procedure
apt-get install cuda

For some reason, apt-get update does not pick up this package from the nvidia repository, so apt-get treats it as a missing dependence under Mint 17.2. The network installation works fine under Mint 17.1, so it looks like this is a bug in Mint 17.2.

An alternative is to use the local, rather than the network install package.

For me, it was a missing python-xkit library.

http://www.ubuntuupdates.org/package/core/trusty/universe/base/python-xkit

after installing that, i was able to install screen-resolution
http://security.ubuntu.com/ubuntu/pool/main/s/screen-resolution-extra/screen-resolution-extra_0.14ubuntu2.1_all.deb

then apt-get update, apt-get install cuda worked.

@rg119 You fixed most of the issues, thanks!

I’m getting issues when trying to run the tensorflow mnist model. Exact error is:

tensorflow/stream_executor/cuda/cuda_dnn.cc:204] could not find cudnnCreate in cudnn DSO; dlerror: /usr/local/lib/python2.7/dist-packages/tensorflow/python/_pywrap_tensorflow.so: undefined symbol: cudnnCreate

Anybody here have any suggestions?

Thanks in advance